Tokens are pretty much the gas mileage. Every time you send a message and Claude answers, you're burning a little gas.

How do tokens relate to my Claude plan?

When you pay for a plan, you get a certain amount of usage over a period. Bigger engines, like Opus, burn more gas per job than smaller ones like Haiku, so the model you pick changes how far your plan goes. See Claude models.

How do tokens relate to the context window?

Each chat has a fixed tank. That's the context window. As a conversation gets long, the tank fills, and Claude starts compacting the chat, keeping only what it thinks matters.

How do I use fewer tokens?

  • Start with a cheaper model and upgrade when the job needs it
  • Start fresh chats instead of running one forever
  • Put the information Claude needs in project files and skills, so you're not pasting it again every time